failure classification in high concentration photovoltaic ... · neural networks, failure analysis....
TRANSCRIPT
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16039
Failure Classification in High Concentration Photovoltaic System (HCPV)
by using Probabilistic Neural Networks
G. Lo Sciuto1, G. Capizzi2 , A. Caramagna3, Fabio Famoso4, R. Lanzafame5 and Marcin Woźniak6
1Department of Electrical, Electronics, and Informatics Engineering, University of Catania Viale A. Doria 6-95125, Catania, Italy.
2Department of Electrical, Electronics, and Informatics Engineering, University of Catania Viale A. Doria 6-95125, Catania, Italy.
3 eFM, Via Laurentina, 455- 00142 Roma, Italy.
4Department of Civil Engineering and Architecture, University of Catania, Italy
Viale A. Doria 6-95125, Catania, Italy. Orcid: 0000-0003-3880-0292 & Scopus ID: 56029382400
5epartment of Civil Engineering and Architecture, University of Catania, Italy
Viale A. Doria 6-95125, Catania, Italy.
6Institute of Mathematics, Silesian University of Technology, Kaszubska 23, 44-101 Gliwice, Poland.
Abstract
In this paper, we present an expert diagnostic system for the
interpretation of four different categories of system’s
functioning based on an innovative feature extraction
tequiniques and a Probabilistic Neural Network for the
classification of events identifying failures that can occur
during in a high-concentration photovoltaic (HCPV) system
located in Fleri, Sicily (Italy). In this paper we have considered
four different categories of system’s functioning: sun tracking
system malfunction, cloudy conditions and temperature sensor
malfunction, darkness and night time, normal functioning.
Keywords: photovolatic systems, performance indicators,
neural networks, failure analysis.
INTRODUCTION
In the last decades the growing attention significance to envi-
ronmental issues has induced many country to investigate the
renewable energy sources as a concrete alternative for the
thermos-mechanical conversion [1, 2]. As a matter of the fact
in last years solar energy has played a key role for innovative
energy systems [3]. Among all possible technologies involved
in the use of solar energy, solar photovoltaic energy is proba-
bly considered one of the most diffuse source of electricity [4].
These technology presents several advantages such as: a
decreasing cost of installed kWp with consequent shorter
payback periods, it can be easily integrated in buildings, no
greenhouse gases emissions [5, 6], a good efficiency in
comparison with other renewable energy technologies [7].
Moreover, PV technology is still the fastest growing power
generation technology [8, 9]. In last decade it has faced an
increase of about 55% each year. As an example, the
worldwide cumulative installed capacity passed from 2,0 GW
in 2006 to about 3,1 GW.
Despite of these considerations the Solar panel manufacturers
are still attempting to make solar panels robust in such way as
to ensure two decades at least. However, the Real Lifespan of
Solar Panels of 20-25 years without maintenance and
degradation, is not real, since the panels subjected to solar
constant exposition, are affected by aging [10], deterioration
of components [11], mechanical and electrical faults such as a
short-circuit failure (water can seep through the surface) or
interruption [12]. The other types of faults are related to exter-
nal phenomena such as atmospheric conditions [13], weather
and wind, and physical damage (e.g. surface scratches for
trees, snow, leaves, twigs and bushes) that cannot be foreseen
or predicted or even removed from the entire photovoltaic
system. The infrequent event is the risk of the lightning
damages to the entire solar system and electronic equipment
[14], on this matter, as deterrent, the lightning rods and cages
are installed according to European Lightning Protection
Standards IEC/EN 62305-1/4. The most important part of the
photovoltaic system is the inverter, performing the conversion
of the power produced used for the household appliances. This
requires constantly to ensure the regular maintenance and
cleaning from dust and any deposits, the replacement of broken
or burned fuses [15]. For all these reasons a great attention is
turned to mathematical models that are able to foresee and treat
mechanical, physical or other types of faults [16, 17]. In this
paper it was investigated a new method to recognize and treat
failures of a new technology with an advanced optical system
to focus the sun light onto each cell for maximum efficiency as
the Concentrating Photovoltaic (CPV) system. From the
analysis of the collected data for HCPV installed different
problems are recognized:
The mechanical arrangement of the sun tracking
system such as the stepper motor can no longer
control the angle of the panel and the rotation of
platform moving the HCPV at different direction of
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16040
solar radiation. In such case the light sensors can
malfunction however the temperature is correctly
detected by the sensor;
The PV module temperature sensor remains failure for
a certain period of time until the maintenance task;
Failure of data communication system;
Any exchanged data for absence of network;
Others problems.
During the night time or darkness the PV panels are not
generally energized and the power generation is not produced.
In case of snow, rain, cloudy weather and dark climate the
solar panel performance decreases.
The method proposed is based on Probabilistic Neural
Network (PNN) for the classification of events that identifying
the failures that can occur in the high-concentration
photovoltaic (HCPV) system located in Fleri, Sicily (Italy).
THE PROBLEM OF FAULTS PREDICTION
Protection is a critical point in Smart Grid and the various
distributed generators connections to distribution feeders
require advanced protection devices to improve security and
reliability. Communication based relays, automatic fault lo-
cation, and automatic sensitivity and selectivity changes in the
protection devices, are characteristics of smart protective
applications to distribution networks [18, 19].
Advanced fault management is also a critical task in Smart
Grid, In fact combining communication-based adaptive protec-
tion relays, wide area monitoring such as phasor measurement
units, and fast acting switching devices with communication
capabilities, an advanced fault management can limit the extent
of an abnormal condition and rapidly restore large portion of
the network, isolating the faulted sections.
The presence of distributed generators in a Smart Grid
environment influences the design of the protection system.
Conventional distribution networks which have unidirectional
current flow employ over-current protection schemes. But,
with the inclusion of distributed generators, network power
flows may change, making conventional protection schemes
unsuitable. Moreover, as the Smart Grid acts as a single entity
to the main grid, the protection scheme should include the
capability to isolate itself in case of a fault in the main grid.
Similarly, the protection scheme should be capable of
localizing faults within the Smart Grid. In particular, the
sensitivity and selectivity of the protection system needs to be
carefully designed to ensure secure and reliable operation of
the Smart Grid. Conventional techniques involve complicated
calculations and may introduce errors in the estimated fault
prediction. These can be overcome by the application of Neural
Networks (NN). We feel that NN are suitable for an efficient
faults prediction [20-22].
Figure 2: The HCPV system installed in Fleri (Sicily-Italy).
Table I: The HCPV Manufacturer’s Specifications
Number of
Modules
8 (with a nominal
peak power of 440 W)
Geometrical
concentration factor 476x
Module's surface
Active collector
area
Weight
Width
Length
Overall efficiency
18 m2
16 m2
930 Kg
8.6 m
8.4 m
25%
The incipient faults in can be detected using diagnostic
techniques by observing the degradation in system
performance. Due to the nature of observed data and available
knowledge, diagnostic methods are often a combination of
statistical inference and machine learning methods [23]. NN
are a better option for diagnosing faults in electrical equipment
for the following reasons:
They can interpolate from previous learnings and give
a more accurate response to unseen data, making them
better at handling uncertainty.
They are fault tolerant, so they handle corrupt or
missing data more effectively.
They are good non-linear function approximators by
nature, making them better at equipment diagnostics.
They are more suitable for extracting the relationship
between input and output in fault detection and
diagnosis applications.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16041
Figure 3: The modules used in the HCPV installed in Fleri.
Table II: Module Parameters.
Cells Multijunction III-V
Number 144
Dimensions
Weight
Certification
Cooling system
Module Power
VOC
ISC
VMPP
IMPP
5.5x5.5 mm
60 kg
IEC- 62108
Natural Conention
440 W
440 V
1.29 A
400 V
1.1 A
THE CONCENTRATING PHOTOVOLTAIC SYSTEMS
(CPV)
The high cost of modules technology, made up of Si, GaAs or
other semiconductors, is leading cell photovoltaic
manufacturers to find new solutions with higher efficiency
values. CPV systems are a concrete answer to this new
demand, by concentrating the sunlight by optical devices like
lenses or mirrors reduces the area of expensive solar cells or
modules, and increases their efficiency. These systems present
the advantage to reduce the active photovoltaic surface used
for the photoelectric conversion exploiting an optical system
that concentrates on it the direct solar radiation. This leads
these systems to reach efficiency values up to almost 50% .
Despite of this main advantage they still present some
disadvantages. The use of only direct solar radiation
component is achieved by a complex system constitute of a
fragile set of components that may incur in frequent faults [m].
Moreover it has been demonstrated that with particular
meteorological conditions, even if they reach higher efficiency
values, the overall yearly energy production may be less than
fixed angle PV systems one. The main components of CPV
systems are usually the following: optical concentrator, heat
sink, sun-tracker, inverter. The concentrator is usually an
optical system constituted by lenses or mirrors with the aim at
concentrating the sun light on the photovoltaic cell. Fresnel
lens is the most used lens in CPV systems. This optical system
could be of three types: point focus when there is one focus for
each cell, linear focus when radiation is focused in one line and
dense-array when there is a focus for groups of closed cells.
Figure 4: Tracker Control Unit (left) and Solarimeter (right).
Figure 5: Aluminum reflective concentrators.
The receiver is constituted of photovoltaic cells, they are
usually multi-junction GE III-V. The IIIV multi- junction solar
cell offers extremely high efficiencies and best response to
high temperatures compared with traditional solar cell made up
of a single layer of semiconductor material.
Multi-junction solar cells can make better use of the solar spec-
trum by having multiple semiconductor layers with different
bandgap energy. Each layer, usually an III-V semiconductor, is
designed to absorb a different portion of the spectrum. By
increasing the number of the spectral regions and thus reducing
the energy intervals of each region, the energy losses through
thermalization may be reduced. The heat sink is necessary for
the systems to keep the temperature relative low and minimize
the losses of energy production due to the typical thermal drift
that usually occurs in CPV systems. The tracking system is
necessary to follow the direct component of the sunlight,
typically it is an electromechanical system at one or two axes
with the aim at keeping the modules always in an horthogonal
position with solar radiation, it has to be accurate especially
with high concentration degrees.
These systems may differ one to each other in function of the
concentration degree of solar radiation. The low concentrating
photovoltaic systems (LCPV) with a concentration degree up
to 40 times uses generally silicon cells with high efficiency.
The high concentrating photovoltaic systems (HCPV) are able
to reach concentration degree up to 500 times with multi-
junction cells.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16042
Table III: Traker Specifications
Azimuth tracking range 0° - 270°
Elevation angle tracking range 0° - 80°
Type
Acceptace half angle
(AHA)
Working maximum wind
speed
Weight
Dual axis
0,3°
90 km/h
450 kg
Table IV: Optical characteristics of Fresnel lens
Primary optics Fresnel lens
Dimensions lens 240x240 mm
Concentration
Focal distance
Secondary optics
476x
200 mm
Reflective
homogenizer
THE CASE STUDY
The HCPV system considered in this paper is installed in Fleri
(Sicily-Italy) (Figure 1), has a nominal peak power of 3,5 kWp
and consists of a mechanical bi-axial tracking system
conteining a matrix of 8 modules, consisting of multi- junction
solar cell on the receiver plate with the Fresnel lens on the top
lens plate, surrounded by a metal frame. The main
characteristics of this system are reported in Table I.
Each module (Figure 2) is constituted of two parts: the optical
group and the tripe-junction cell with its heat sinker. This latter
is located on the back of the module. It is essential to the
problem of thermal lift. Usually, in traditional systems such as
fixed angle PV systems with silicon cells, the thermal power
loss coefficient (percentage variation of power for °C) at
standard test conditions is about 0,40%/°C . The considered
HCPV suffers less and its coefficient is -0,25%/°C. The
module’s parameters are displayed in Table II.
The concentration of irradiation (476 suns) is achieved by the
optical design, using Fresnel lens which has circular symmetry
about its axis and concentrates sunlight onto the GaAs solar
cell. This optical system is point focus, each cell is provide of
its own main and secondary optical lens.
HCPV modules, in contrast to PV, need to be constantly
aligned with respect to the direct light sun beam. This is why
HCPV are mounted on sophisticated trackers which follow the
sun in order to guarantee normal incidence of the direct light
beam. If a concentrator is not perfectly aligned with the sun
beam, it will lose part of the available energy.
Dual axis sun tracking systems are required for HCPV
applications while the tracking system improves the efficiency
of PV system. In this system there is a dual axis tracker that
can follow the sun both vertically and horizontally and thus it
can track the sun’s apparent motion virtually anywhere. The
solar tracking strategy uses optical tracking technique.
Figure 6: Solar irradiance in the module plane for a typical
sunny day (14/05/2012) in Fleri (Sicily-Italy).
Figure 7: AC Power generation of the HCPV for a typical
sunny day (14/05/2012) in Fleri (Sicily-Italy).
These dual axis dries combine the azimuth and elevation
movement.
The drive mechanisms include linear actuators, linear drives,
swivel drives, worm gears, planetary gears, and threaded
spindles. The drive type is a rotary gear drive with a driven
motor power of 2500 rpm/48W/24V DC, mechanism used as
azimuth and linear drive provides the necessary mechanical
movement and torque to enable solar tracking used in elevation
angle (3000 rpm/72W/24V CC).
The movement is controlled both with astronomical equip-
ment and optical system. The astronomical equipment uses
astronomical equations of the movement of earth around the
sun and is not affected by the presence of clouds. The optical
system takes advantage of sensors present in the system to
follow the apparent movement of sun but suffers of the
presence of clouds (Figure 3).
The dual axis sun tracking systems is used to optimize and
reduce the tilt angle. This misalignment angle at which the
performance of concentrator is reduced by more than 10% is
called acceptance angle. The maximum acceptance angle for
the HCPV under consideration is 0, 3°. The specifications of
the tracking system are shown in Table III.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16043
Figure 6: Top diagrams display the mean and reactive power during a windowed interval of time in normal operating conditions
and its windowed Fourier transform while bottom diagrams display the mean and reactive power during a windowed interval of
time in a faulty condition and its windowed Fourier transform.
The optical system is composed of two parts: a primary optic
and a secondary one. The main optic is constitute of Fresnel
lenses made in PMMA. The efficiency of these lenses is about
90%. It suffers usually of a deflection of the light beam caused
by the not uniform polymers. The secondary optics is a
reflective homogenizer that distributes homogenously the
radiation concentrated by the lens on the whole active surface
(in Figure 4 are shown the reflectors cones). In Table 4 are
reported the main characteristics of the optical system.
The photovoltaic cells, used in the HCPV, are triple-junction
III-V Ge with an efficiency of 39%. The combination of these
cells with a point focus Fresnel lens (IEC 62108) with a
concentration of 476 x allows to reach an overall efficiency of
about 25%. The triple-junction allows to convert the
photoelectric radiation to electrical energy with a more
extended range of wave lengths (300-1700 nm) in comparison
with the typical cell of silicon (300-1140 nm).
EXPERIMENTAL DATA
The data have been collected at 15-minute intervals, in
particular, the irradiation (kW/m2), the modules temperature
(°C), the inverter’s current (A), the produced energy (kWh)
and the active power (kW). The data is recorded during the
period starting from January 2014 to December 2014.
In Figure 6 and 7 there are reported the direct radiation in the
modules and the power calculated during a sunny day.
In order to analyze the system performances it was used the
Performance ratio (PR). The performance ratio is calculated
by:
𝑃𝑅 =𝑌𝐹
𝑌𝑅
Where:
The Array Yield (YF) is the ratio between the energy
production E [kWh] and the peak power of the system
P0 [kWp]. It normalizes the energy produced with
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16044
respect to the system size, being a convenient way to
compare the energy produced by systems of differing
size;
- The Reference Yield (YR) is the ratio between the
solar radiation directed perpendicularly to the active
surface (EDNI) [kWh/m2] and the Direct Normal
Irradiance (DNI) (i.e. 850 [W/m2] for concentrated PV
systems). It represents the number of working hours at
the DNI considered.
PR compares the real production of the system with the
attended one. Moreover, it is a good performances index
because it shows the proportion of the energy loss (e.g. due to
thermal losses and conduction losses, inverter inefficiency,
wiring, mismatch and other losses when converting from D.C.
to A.C power) and of energy consumption for operation. The
theoretical power in a HCPV system is calculated as:
𝑃𝐻𝐶𝑃𝑉 = 𝑃𝐻𝐶𝑃𝑉 ∙ 𝜎 ∙ 𝐶 ∙ 𝜂𝑆 ∙ 𝐴𝐻𝐶𝑃𝑉 ∙ 𝜂𝑆 (1)
Where
- IHCPV is the direct component of solar radiation;
- σ ∗ C is concentration coefficient net of optical losses;
- AHCPV is the photovoltaic surface;
- ηs and ηi are the efficiency of the module and the
inverter.
In this paper we have considered four different categories of
system’s functioning. More specifically :
1. Sun tracking system malfunction ;
2. Cloudy conditions and temperature sensor
malfunction;
3. Darkness and night time;
4. Normal functioning.
METHODOLOGY
The conventional way of modeling of the algorithms for the
failures detection and classification needs to be augmented or
replaced with intelligent techniques that are robust and fault-
tolerant. The Neural Networks (NN) provides a promising
alternative in the situations outlined above [24, 25].
One of the primary objectives of the renewable sources is to
operate economically and reliably. This depends greatly on
accurate demand forecasting and renewable generation
forecasting. The purpose of demand forecasts is to predict the
power that will be consumed by the load. In conventional
power systems, demand depends on the weather, type of day
and random activities or incidents.
There are various techniques based on statistic and on arti-
ficial intelligence which can forecast the load for the next 24
hours with a good degree of accuracy. Due to high variations in
the load profile, it is difficult to accurately forecast demand.
Hence, the models developed should try to predict the load
with reasonable accuracy. Some of the factors to be considered
while developing the NN model are:
Increased responsiveness of demand to predicted real-
time electricity prices as compared to conventional
grids.
Classification of load profiles of different customer
classes based on the consumption of electricity.
Need to input past demand as different components
(like base load, peak load, valley load, average load
and random load), instead of combining them into a
single input.
Type of day (weekday/weekend) input may not have
much effect on forecasting for a Hospital Smart Grid,
whereas it is a major factor impacting the load profile
of a Residential Smart Grid.
In this paper we face the problem of faults prediction by using
an approach based on neural networks and on a new time-
frequency feature extraction technique. An extended statistical
study of the historical data series of reactive power and mean
power generated by the HCPV shown that when a fault
approaches, the reactive power is affected by a series of bad
anomalous effects. It is virtually impossible to determine the
peculiar modification of each singular side effect as much as
their composition, but it is possible to intercept the time
window and the spectral signature that these effects generate
respect to the normal functioning.
Figure 8: The proposed cascade PNN architecture.
The implementation was done in order to obtain the correct
prediction of what should be the time serie progression relative
to a windowed interval of time, and, simultaneously, intercept
the spectral characteristics of such predicted progression to
identify an overcoming fault.
In order to characterizing the spectral signature we have
calculated the expected and variance values of the mean and
reactive power both in the time domain and in the Fourier
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16045
domain these calculation are performed by using a sliding
window. The time amplitude of the window is 2,5 hours.
In Fig. 5 is shown a sample of the expected and variance
values of the mean and reactive power in regular and faulted
conditions (Sun tracking system malfunction).
VII. THE PROPOSED NEURAL NETWORK ARCHIECTURE
The PNN implementated in this paper is organized as
multilayered feedforward network with four layers:
Input layer.
Pattern layer.
Summation layer.
Output layer
The PNN is created by a set of multivariate probability
densities which are generated from the training vectors
presented to the PNN. The input instance with unknown
category is propagated to the pattern layer. The outputs are a
linear combination of the hidden nodes outputs. More
specifically, the k-th network output has the form [26]:
𝑦𝑘(𝑥(𝑡)) = ∑ 𝑤𝑘,𝑗𝑀𝑗=1 𝜙𝑗(𝑥(𝑡)) (2)
where
𝜙𝑗(𝑥(𝑡)) = 𝑒𝑥𝑝 {−1
2𝜎𝑗(𝑥(𝑡) − 𝜇𝑗)
𝑇Σ𝑗−1(𝑥(𝑡) − 𝜇𝑗)} (3)
μj and Σj are the function center (mean vector) and covariance
matrix of the j-th basis function respectively and σj is a
smoothing parameter controlling the spread of the j-th basis
function.
RESULTS AND DISCUSSION
To evaluate the pattern recognition algorithm, dataset is
randomly split into three parts: a training set consisting of 900
data points (600 data points representative of the various kind
of faults and 100 representative of normal operative condition)
a validation set consisting of 100 data points and a testing set
consisting of 100 data points. The training set is used to find
the model parameters in the used PNN network. These
parameters are the number of neuron for each class and the the
weights value.
The resulting network after the training phase and pruning is
shown in Fig. 8. It consists of 120 neurons (40 for each class).
Once the optimal parameters are found the trained algorithm is
applied to classify the data points in the testing dataset into one
of the two classes. A correct classification rate of 90% average
has been obtained.
CONCLUSION
In this study, we developed an expert diagnostic system for the
interpretation of four different categories of system’s
functioning by using an innovative feature extraction
tequiniques and a probabilistic neural networks (PNN). More
specifically :
1. Sun tracking system malfunction.
2. Cloudy conditions and temperature sensor malfunc-
tion.
3. Darkness and night time.
4. Normal functioning.
The stated results show that the proposed method can make an
efficient interpretation with an accuracy of 90\%.
REFERENCES
[1] Dresselhaus M., Thomas I., 2001, “Alternative energy
technologies,” Nature, 414, no. 6861, pp. 332.
[2] Farret F., Simoes M.G., 2006, “Integration of alternative
sources of energy.” John Wiley & Sons.
[3] A. S. Joshi A.S., Dincer I., aReddy B.V., 2009,
“Performance analysis of photovoltaic systems: A
review,” Renewable and Sustainable Energy Reviews,
13(8), pp. 1884 – 1897.
[4] Branker K., Pathak M., Pearce J.M., 2011 “A review of
solar photo- voltaic levelized cost of electricity,”
Renewable and Sustainable Energy Reviews, vol. 15(9),
pp. 4470–4482.
[5] Fthenakis V.M., Kim H.C., Alsema E., 2008,
“Emissions from photo- voltaic life cycles,”
Environmental science & technology, vol. 42(6), pp.
2168–2174.
[6] Alsema E., 2000, “Energy pay-back time and co2
emissions of pv systems,” Progress in photovoltaics:
research and applications, vol. 8(1), pp. 17–25.
[7] El Chaar L., El Zein N., 2011, “Review of photovoltaic
technologies,” Renewable and sustainable energy
reviews, vol. 15(5), pp. 2165– 2175.
[8] F. Dinc ̧er F., 2011, “Overview of the photovoltaic
technology status and perspec- tive in turkey,”
Renewable and Sustainable Energy Reviews, 15(8), pp.
3768–3779.
[9] Bhutto A.W., Bazmi A.A., Zahedi G., 2012, “Greener
energy: issues and challenges for pakistansolar energy
prospective,” Renewable and Sustainable Energy
Reviews, vol. 16(5), pp. 2762–2780.
[10] Famoso F., Chiacchio F., Brusca S., D’Urso D., 2016,
“Stochastic hybrid fault tree automaton for the
production forecast of pv power plant,” , Risk,
Reliability and Safety: Innovating Theory and Practice,
pp.824-830.
[11] Garcia M., Marroyo L., Lorenzo E., Marcos J., Perez
M., 2014, “Observed degradation in photovoltaic plants
affected by hot-spots,” Progress in Photovoltaics:
Research and applications, vol. 22(12), pp. 1292– 1301.
[12] Reil F., Sepanski A., Herrmann W., Althaus J., Vaaßen
W., Schmidt H., 2012, “Qualification of arcing risks in
pv modules,” in Photo- voltaic Specialists Conference
(PVSC), 38th IEEE. IEEE, 2012, pp. 000 727–000 730.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 24 (2017) pp. 16039-16046
© Research India Publications. http://www.ripublication.com
16046
[13] Armstrong S., Hurley W., 2010, “A thermal model for
photovoltaic panels under varying atmospheric
conditions,” Applied Thermal Engineering, 30(11), pp.
1488–1495.
[14] Hernandez J.C., Vidal P.G., Jurado F., 2009, “Lightning
and surge protection in photovoltaic installations,” IEEE
Transactions on power delivery, 23(4), pp. 1961–1971.
[15] Rao K.U., Parvatikar A.G., Gokul S., Nitish N., Rao P.,
2016, “A novel fault diagnostic strategy for pv micro
grid to achieve reliability centered maintenance,” in
Power Electronics, Intelligent Control and Energy
Systems (ICPEICES), IEEE International Conference
on. IEEE, pp. 1–4.
[16] Chao K.H., Ho S.H., Wang M.H., 2008, “Modeling and
fault diagnosis of a photovoltaic system,” Electric
Power Systems Research, 78(1), pp. 97–105.
[17] Firth S.K., Lomas K.J., Rees S.J., 2010, “A simple
model of pv system performance and its use in fault
detection,” Solar Energy, 84(4), pp. 624–635.
[18] G. Capizzi G.,Sciuto G.L., Napoli C., Tramontana E.,
"Advanced and Adaptive Dispatch for Smart Grids by
means of Predictive Models," IEEE Transactions on
Smart Grid, 99, pp. 1-1.doi:
10.1109/TSG.2017.2718241.
[19] Capizzi G., Lo Sciuto G. L., Napoli C., Tramontana E.,
2017, “An advanced neural network based solution to
enforce dispatch continuity in smart grids,” Applied Soft
Computing, https://doi.org/10.1016/j.asoc.2017.08.057.
[20] Schetinin V., 2003, “A learning algorithm for evolving
cascade neural networks,” Neural Processing Letters,
17(1), pp. 21-23.
[21] Yan W., 2012, “Toward automatic time series
forecasting using neural net- works,” IEEE Trans. on
Neural Networks and Learning Systems, 23(7), pp.
1028-1039.
[22] Bonanno F.,Capizzi G.,Gagliano G., Napoli C., 2012,
“Optimal Management of Various Renewable Energy
Sources by a New Forecasting Method,” 2012
International Symposium on Power Electronics
Electrical Drives Automation and Motion (SPEEDAM),
pp. 934-940, 20-22.
[23] Borges R.V.,Avila Garcez A.D., Lamb L.C., 2011,
“Learning and Rep- resenting Temporal Knowledge in
Recurrent Neural Networks,” IEEE Transactions on
Neural Networks,, 22 (12), pp. 2409-2421.
[24] Brusca S., Capizzi G., Lo Sciuto G., Susi G., 2017, “A
new design methodology to predict wind farm energy
production by means of a spiking neural network-based
system.” Int J Numer Model. 2017,
e2267.https://doi.org/10.1002/jnm.2267.
[25] G. Lo Sciuto G. et al., 2017, “Combining SVD and Co-
occurrence Matrix Information to Recognize Organic
Solar Cells Defects with a Elliptical Basis Function
Network Classifier.” Rutkowski L., Korytkowski M.,
Scherer R., Tadeusiewicz R., Zadeh L., Zurada J. (eds)
Artificial Intelligence and Soft Computing. ICAISC
2017. Lecture Notes in Computer Science, vol 10246.
Springer, Cham.
[26] C. M. Bishop C. M., 2006, “Pattern Recognition and
Machine Learning,” Springer.